A new approach for the domain-specific opinion word extraction is proposed. This approach is based on several text collections and various statistical features computed using them. The extracted opinion words are used in the three-way review classification problem. In this problem, the reviews are divided into the following groups: "thumbs up", "so-so", and "thumbs down". In order to solve this problem, we use various features, such as opinion words, word weights, punctuation marks, and operator words that can affect the polarity of the next words.

Keywords: knowledge acquisition, opinion word extraction, review classification, machine learning

• PostScript (in Russian) (784KB)
• PDF (in Russian) (219KB)



• PostScript. zip (in Russian) (395KB)
• PDF. zip (in Russian) (192KB)